c67910fa6f
* bump hf deps * upgrade liger-kernel too * install cce from fork for transformers fix * fix reference to vocab size in gemma3 patch * use padding_idx instead of pad_token_id * remove fixed gemma3 patch * use updated cce fork * fix local mllama cce patches w docstring * add test for multipack with trainer setup and fix trainer for trainer refactor upstream * bump modal version * guard for iterable datasetS * mllama model arch layout changed in latest transformers * fix batch sampler with drop_last * fix: address upstream vlm changes for lora * fix: update references to old lora target path * fix: remove mllama fa2 patch due to upstream fix * fix: lora kernel patch path for multimodal models * fix: removed mllama from quarto * run test for came optim on 2.6.0+ * fix fsdp2 patch and remove deprecated patch * make sure to set sequence_parallel_degree for grpo * Add SP test for GRPO * add sp to grpo config for trainer * use reward_funcs as kwarg to grpo trainer * fix the comprehension for reward funcs * reward funcs already passed in as args * init sp_group right before training * fix check for adding models to SP context * make sure to pass args to super * upgrade deepspeed * use updated trl and add reasoning flags for vllm * patch the worker --------- Co-authored-by: NanoCode012 <nano@axolotl.ai>
250 lines
8.2 KiB
Python
250 lines
8.2 KiB
Python
"""
|
|
E2E tests for custom optimizers using Llama
|
|
"""
|
|
|
|
import unittest
|
|
|
|
from axolotl.cli.args import TrainerCliArgs
|
|
from axolotl.common.datasets import load_datasets
|
|
from axolotl.train import train
|
|
from axolotl.utils.config import normalize_config, validate_config
|
|
from axolotl.utils.dict import DictDefault
|
|
|
|
from .utils import (
|
|
check_model_output_exists,
|
|
require_torch_2_5_1,
|
|
require_torch_2_6_0,
|
|
with_temp_dir,
|
|
)
|
|
|
|
|
|
class TestCustomOptimizers(unittest.TestCase):
|
|
"""
|
|
Test case for Llama models using LoRA
|
|
"""
|
|
|
|
@with_temp_dir
|
|
def test_optimi_adamw(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.02,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "optimi_adamw",
|
|
"max_steps": 5,
|
|
"lr_scheduler": "cosine",
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert trainer.optimizer.optimizer.__class__.__name__ == "AdamW"
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_5_1
|
|
def test_adopt_adamw(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.02,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"max_steps": 5,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "adopt_adamw",
|
|
"lr_scheduler": "cosine",
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert "ADOPT" in trainer.optimizer.optimizer.__class__.__name__
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_5_1
|
|
def test_muon(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.02,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"max_steps": 5,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "muon",
|
|
"lr_scheduler": "cosine",
|
|
"weight_decay": 0.01,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert "Muon" in trainer.optimizer.optimizer.__class__.__name__
|
|
|
|
@with_temp_dir
|
|
def test_fft_schedule_free_adamw(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"sequence_len": 1024,
|
|
"val_set_size": 0.01,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"micro_batch_size": 2,
|
|
"gradient_accumulation_steps": 2,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "schedule_free_adamw",
|
|
"lr_scheduler": "constant",
|
|
"save_safetensors": True,
|
|
"max_steps": 10,
|
|
}
|
|
)
|
|
# pylint: disable=duplicate-code
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_6_0
|
|
def test_came_pytorch(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "JackFram/llama-68m",
|
|
"tokenizer_type": "LlamaTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.1,
|
|
"special_tokens": {
|
|
"unk_token": "<unk>",
|
|
"bos_token": "<s>",
|
|
"eos_token": "</s>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "came_pytorch",
|
|
"adam_beta3": 0.9999,
|
|
"adam_epsilon2": 1e-16,
|
|
"max_steps": 5,
|
|
"lr_scheduler": "cosine",
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|